Unique Research · Insight / Unique Friend Q&A · January 21, 2024 · Guest: Zhou Jian, founder, Lanma Tech (澜码科技)
Inventory type remains insight (Unique Friend series — not partner welcome).
Guest
Zhou Jian — BS/MS Computer Science, Shanghai Jiao Tong University; 2002 ACM ICPC World Champion (first Asian team member to win). Joined Google HQ in 2006 for Chinese web search quality; later R&D director / CTO roles at Alibaba Cloud, MediaV, Yitu Tech, and Laiye RPA. Serial founder for a decade — employee #10 at Yitu, CTO at Laiye — with deep AI, big-data, and enterprise-services experience.
Lanma Tech builds an LLM-based Agent platform; core team from Google, IBM, Tencent, ByteDance, Alibaba, Yitu and peers. Its enterprise AI Agent platform AskXBOT connects people and systems so firms can automate on expert knowledge. Lanma has closed a tens-of-millions-RMB Series A with IDG Capital, Lianxin Capital, and Atom Capital, and partners with listed and unicorn companies.
Q&A
1. Product and core scenes?
AskXBOT is Lanma’s one-stop LLM Agent platform for design, development, use, management, and knowledge capture. Enterprise users describe needs in conversation, then design, create, and manage Agents that complete tasks with higher quality and lower cost. Four modules: Agent Designer, Knowledge Center, end-user surface, and admin console.
2. R&D / go-to-market in 2023 — biggest challenge and win?
Problem–Solution Fit is done; Product–Market Fit is underway in finance, energy, and retail. Hardest: mapping the capability boundary of foundation models and riding a fast-changing model ecosystem — Agent product design must rest on correct judgment of capability and cost; building on shifting sand is hard. Biggest win: new AI software is not “software” but work — it replaces tasks/skills, not jobs. That view is easy; reproducible practice is what shapes the product roadmap.
3. Opportunities vs. past waves?
GenAI / LLMs turn compute into intelligence and create a new Moore’s Law — inference cost falls ~10× every 18 months. Supply of productivity will grow exponentially and force demand-side change. With industrial and consumer internet already infrastructure, any scene breakthrough scales society-wide — ChatGPT’s path to 100M DAU is the emblem. Difference vs. past: small teams can amplify ideas into exponential products on model infrastructure — but building a durable moat is harder than ever.
4. Challenges landing AI in China?
In ToB: scarce compute; uneven customer understanding of AI, so sales cycles and expectation alignment burn time.
5. Keys for product and GTM?
Respond fast to model change in product and tech. Markets are layered and at different cycle stages — judgment in execution must be fine-grained.
6. Competing with giants’ “dimensionality reduction”?
Startups must poke a hole with an embroidery needle. Giants never apply enough pressure on a single point; go deep on one scene.
7. Are LLMs still too weak for products?
The leading model will always feel “not enough.” Uneven model capability is exactly where opportunity lives — if one model were both cheap and strong, why would founders be needed?
8. Human–machine collaboration?
Models make machines adapt to people. Only a few apps will earn direct co-work with humans. Enterprises should start with RAG, feed more knowledge into models, then self-serve analytics to aid decisions.
9. 2024 tech to watch?
Multimodal and embodied intelligence.
1 CN image; attach on ingest. Signup CTA in CN omitted.
— END —